LLM Stats (OSWorld): leaderboard
369 computer tasks in real Ubuntu/Windows/macOS environments testing autonomous GUI interaction, file management, and application use.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seed 2.1 Pro | 78.8 | #67 |
| 2 | Seed 2.1 Turbo | 76.4 | #86 |
| 3 | Claude Opus 4.6 | 72.7 | #63 |
| 4 | Claude Sonnet 4.6 | 72.5 | #90 |
| 5 | Qwen 3 VL 235B A22B Instruct | 66.7 | #273 |
| 6 | Claude Opus 4.5 | 66.3 | #83 |
| 7 | GLM-5V Turbo | 62.3 | #166 |
| 8 | Seed 1.8 | 61.9 | #139 |
| 9 | Claude Sonnet 4.5 | 61.4 | #142 |
| 10 | Claude Haiku 4.5 | 50.7 | #280 |
| 11 | Qwen 3 VL 32B (Thinking) | 41 | #294 (Qwen 3 VL 32B) |
| 12 | Qwen 3 VL 235B A22B (Thinking) | 38.1 | #238 (Qwen 3 VL 235B A22B) |
| 13 | Qwen 3 VL 8B Instruct | 33.9 | #420 |
| 14 | Qwen 3 VL 8B (Thinking) | 33.9 | |
| 15 | Qwen 3 VL 32B Instruct | 32.6 | #283 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-osworld · How It Works · Data refreshed daily, snapshot 2026-10-20.